TANDARDS COMPETITION IN THE PRESENCE OF …ckemerer/standards-competition-misq.pdf · importantly for our study—digital flash memory cards (e.g., Compact Flash, SmartMedia, Secure

RESEARCH ARTICLE

STANDARDS COMPETITION IN THE PRESENCE OF DIGITALCONVERSION TECHNOLOGY: AN EMPIRICAL ANALYSIS

OF THE FLASH MEMORY CARD MARKET1

Charles Zhechao LiuUniversity of Texas at San Antonio, San Antonio, TX 78249 U.S.A. {[email protected]}

Chris F. KemererUniversity of Pittsburgh, Pittsburgh, PA 15260 U.S.A. and

King Abdul Aziz University, Jeddah, SAUDI ARABIA {[email protected]}

Sandara A. SlaughterGeorgia Institute of Technology, Atlanta, GA 30308 U.S.A. {[email protected]}

Michael D. SmithCarnegie Mellon University, Pittsburgh, PA 15213 U.S.A. {[email protected]}

Both theoretical and empirical evidence suggest that, in many markets with standards competition, networkeffects make the strong grow stronger and can “tip” the market toward a single, winner-take-all standard. Wehypothesize, however, that low cost digital conversion technologies, which facilitate easy compatibility acrosscompeting standards, may reduce the strength of these network effects. We empirically test our hypotheses inthe context of the digital flash memory card market.

We first test for the presence of network effects in this market and find that network effects, as measured here,are associated with a significant positive price premium for leading flash memory card formats. We then findthat the availability of digital converters reduces the price premium of the leading flash card formats andreduces the overall concentration in the flash memory market. Thus, our results suggest that, in the presenceof low cost conversion technologies and digital content, the probability of market dominance can be lessenedto the point where multiple, otherwise incompatible, standards are viable.

Our conclusion that the presence of converters weakens network effects implies that producers of non-dominantdigital goods standards benefit from the provision of conversion technology. Our analysis thus aids managersseeking to understand the impact of converters on market outcomes, and contributes to the existing literatureon network effects by providing new insights into how conversion technologies can affect pricing strategies inthese increasingly important digital settings.

1

Keywords: Network effects, network externalities, standards competition, conversion technologies, flashmemory, digital goods, market competition

1Vijay Gurbaxani was the accepting senior editor for this paper. Bruce Weber served as the associate editor.

MIS Quarterly Vol. X No. X, pp. 1-XX/Forthcoming 2012–2013 1

Liu et al./Standards Competition in the Presence of Digital Conversion Technology

Introduction

Network effects arise in many information technology (IT)markets where the value of a product or service to oneconsumer is at least partially dependent on the choices madeby other consumers. These direct interoperability benefitsmake the choice of a technology standard or platform animportant strategic decision for both consumers and firms(Economides 1996; Katz and Shapiro 1985). Examples ofinformation technologies with demonstrated network effectsinclude computer hardware (Chen and Forman 2006),operating systems (Bresnahan 2001), application software(Brynjolfsson and Kemerer 1996; Gallaugher and Wang 2002;Gandal 1994), and popular instant messaging and socialnetworks (Sundararajan 2007).

In addition to direct network benefits, a widely adopted pro-duct may indirectly give rise to a longer product life cycle,better product support and services, and a greater variety ofcomplementary goods. Given these direct and indirect bene-fits, consumers are more likely to choose products that adoptmore popular standards. This behavior can, in turn, create avirtuous cycle for the leading (majority) formats and helps thestrong grow stronger (Shapiro and Varian 1999). This type ofmarket evolution has been documented in the VHS andBetamax “standards competition” (Cusumano et al. 1992;Park 2004), the adoption of the DVD format (Dranove andGandal 2003), and in the markets for U.S. desktop operatingsystems and office productivity software (Bresnahan 2001;Brynjolfsson and Kemerer 1996).

However, in the past decade, a new and different pattern ofcompetition seems to be emerging in several IT markets withdigital data storage and exchange. Despite strong demand forcompatibility, these markets have, to date, not clearly tippedtoward a single standard, nor do we see a significant advan-tage for the incumbent over the new entrants. Witness, forexample, markets for digital media files (e.g., Real Media,Windows Media, QuickTime, AVI, and MPEG), digitalimage files (e.g., JPEG, GIF, TIFF, and PNG), and—importantly for our study—digital flash memory cards (e.g.,Compact Flash, SmartMedia, Secure Digital, Memory Stick,XD Picture, and Multimedia). Why are the competitivedynamics in these digital goods markets seemingly differentfrom the “tipping” markets mentioned earlier?

In this study, we investigate this question in the context of theflash memory card market. Flash memory is a class of non-volatile, electrically rewritable memory that was introducedinto the consumer electronic market in the 1990s.2 With the

capability to store large amounts of data in a digital format,fast read/write speeds, and compact size, flash memory hasemerged as the primary storage media of various digitalelectronic devices such as digital cameras, digital camcorders,mobile phones, PDAs, and audio players. The popularity ofthese platforms has made the flash memory card market oneof the fastest growing sectors in the IT industry.3 This growthmay not be surprising given the rapid expansion of consumerelectronics devices using flash memory for data storage andtransfer. However, what is surprising is the variety of distinctcard formats that exist in the market in spite of apparent directand indirect network effects.

Direct network effects may arise from an individual userdesiring to transfer, edit, or share media content acrossvarious digital devices that she owns, or from the desire toshare digital content between different users.4 These sorts ofdirect network benefits are commonly discussed in theindustry literature and in the popular press. For example,Sony has touted the ability to use the Memory Stick format on“an extensive range of compliant products,” including thepossibility that “photos taken during a trip can be viewedimmediately on your [Memory Stick compatible] navigationsystem’s large screen monitor” or printed at an in-store kioskterminal that accepts the Memory Stick format.5 Similarly,SanDisk, another major manufacturer of flash memory cards,promotes the availability of its format on a wide range ofconsumer products, and that if you upgrade between SD com-patible phones you can “Simply save all of the information onthis SD card and move the information to your new phone!”6

Likewise, the Wall Street Journal has reported: “Memory-card selection has increasingly important ramifications forgadget owners. Having a card that you can pop in and out ofdifferent devices lets you do things you couldn’t otherwise.You can, for instance, take a picture on a digital camera,transfer the memory card to your personal organizer, and thenbring up the picture on its screen” (McWilliams 2003).

2http://en.wikipedia.org/wiki/Flash_memory_card.

3http://www.usatoday.com/tech/products/2006-06-04-storage-drive_x.htm.

4See, for example, McWilliams (2003): “Mike Rogers snapped his waythrough France last year, taking heaps of digital pictures at every stop.Whenever he ran out of storage space on his camera, he simply boughtanother memory card to pop in the back. Mr. Rogers is among a growingnumber of Americans who, having loaded up on personal organizers, MP3players and digital cameras, are quickly running out of room to save all theiraddresses, videos, songs and pictures. That is forcing them back into thestores to stock up on extra memory—and fueling a huge new business for theconsumer-electronics industry.”

5http://www.sony.net/Products/memorycard/en_us/memorystick/index.html.

6“SanDisk Launches V-Mate,” Business Wire, September 2, 2006 (http://findarticles.com/p/articles/mi_m0EIN/is_2006_Sept_2/ai_n16702773).

2 MIS Quarterly Vol. X No. X/Forthcoming 2012–2013


There is increasing awareness among (especially younger)consumers that, by adopting the same flash memory cardformat for their collection of digital devices, they can easilyexchange data (e.g., play the video they recorded with adigital camcorder on an HDTV7) and eliminate the need topurchase a variety of different memory card formats (e.g.,family members on vacation can benefit from being able toswap memory cards from one camera to another when onecard becomes full of photographs8).

Given the above, market share may be a critical factor for newconsumers who face their first adoption decision of digitaldevices, as once they have decided to adopt a particular flashmemory format it could be costly to reverse such a decision. To the degree that direct network effects are present in thismarket, consumers who purchased a particular flash memoryformat will be more likely, ceteris paribus, to take this formatinto consideration when purchasing their next digital devicein order to be able to reuse the flash memory cards and toeasily transfer digital outputs from one device to another.These network effects could reasonably extend beyondconsumers’ own digital devices to the format choices of otherfamily members or friends with whom they might exchangedigital data.

In addition to direct network effects, indirect network effectsmay arise in the flash memory card market to the degree that,as one format becomes widely adopted, more manufacturerswould find it desirable to make their devices compatible withan emerging standard for flash memory cards, which wouldgenerate increasing returns to adoption. In this way, themarket for flash memory cards may reflect the accumulationof individual consumer decisions since, as consumersrecognize the advantages of easy digital exchange and flashmemory card reuse, they may tend to purchase cards anddevices that are compatible with cards that they have alreadypurchased. As manufacturers observe this, they would bemore likely to bring products to market that have compatibleslots for the most popular formats, which would createpositive feedback for greater use of those formats.

However, despite vendors’ stated efforts and the acknowl-edgement by the business press of network effects for flashmemory devices, the typical self-reinforcing feedback loop ofa popular format becoming more popular and leading to a

winner-take-all market has not yet occurred in the flashmemory market as might have been predicted.9 As depictedin Figure 1, the flash memory market has been split amongseveral different formats, with little evidence of marketconsolidation during this time period. This lack of consoli-dation has been observed in the popular press in reportingback to consumers, for example, in the New York Times.10

It is also interesting to note that the same phenomenon existsin Europe as well as in the United States. Some Dutchresearchers have conducted a study to investigate why mul-tiple flash memory card formats coexist despite the expecta-tion that network effects would drive the market toward awinner-take-all outcome (de Vries et al. 2009). The re-searchers conducted face-to-face interviews with marketingmanagers from several large flash memory card manufacturers(SanDisk, Sony, and Olympus) and MediaMarket, the largestconsumer electronics retailer in Europe. Based on the inter-views, the researchers argue that both manufacturers andconsumers are aware of the existence of network effects in theflash memory card market and they hypothesize that acombination of factors may result in multiple memory formatscoexisting, including the presence of converters, or what theyterm “gateway technologies.”

The apparent “winners-take-some” market outcome reflectedin Figure 1 raises two main research questions. The firstquestion is: Are network effects present in this market? Theindustry, press reports and academic research mentionedabove document the belief that network effects are present inthe flash memory card market. However, the flash memorycard market may be different from the classic “networkeffects driven” markets such as those for VCRs and faxmachines. Unlike many other technologies commonly studiedin the literature, flash memory cards carry digital content thatcan be transferred easily and losslessly between devices usinga variety of inexpensive PC- and USB-based converters.These converters, or multiformat readers, are not physicaladapters that provide direct interoperability; rather, they areUSB-based hardware devices that have multiple memory cardslots for different types of otherwise incompatible flash

7Business Wire, September 2, 2006 (http://findarticles.com/p/articles/mi_m0EIN/is_2006_Sept_2/ai_n16702773). Also see McWilliams (2003);http://www.ehow.com/how_5220958_transfer-one-wireless-phone-another.html; and http://www.memorystick.com/en/lifestyle/incar/index.html.

8This actually happened to one of the authors.

9Figure 1 shows the market shares of the six major flash memory cardformats (slot interfaces) from 01/2003 to 08/2006. As flash memory cardtechnology has advanced after 2006, newer memory card formats are notdirectly comparable with the prior versions examined in our study and aretherefore not shown here.

10See the New York Times video segment, “Yo, Jude, Memory Cards 101,”http://video.nytimes.com/video/2009/09/28/technology/personaltech/1247463931097/yo-jude-deciphering-memory-cards.html? nl=technology&emc=techupdateema5.

MIS Quarterly Vol. X No. X/Forthcoming 2012–2013 3


Figure 1. Flash Memory Card Monthly Market Share, January 2003 to August 2006 (NPD Group)

The two-way arrows are meant to indicate general data flows across a variety of settings. In specific applications, it is possible that data

may only flow in one direction.

Figure 2. Consumer’s Choice Set Before and After the Introduction of Converters

D3

Before the Introduction of Converters (Arrows indicate the flow of data):

After the Introduction of Converters:

D1

D1

D2 F2

Digital Device (i.e., Smartphone)

Digital Devices

F1

F3

Digital Device Flash Memory Card

D2

D4

Flash Memory Card

F1

Converterand PC

Flash Memory Card

F4

F2



cards.11 Through a USB cable, consumers can read from orwrite to multiple flash memory cards simultaneously from aPC. Figure 2 illustrates the consumer’s choice set before andafter the introduction of converters. As can be seen inFigure 2, converters expand the consumer’s ability to usedigital devices of different formats by allowing conversionfrom one flash memory format to another. We provide a moredetailed list of “use cases” where consumers might use aconverter to share data across otherwise incompatible flashmemory formats in the Appendix.

Currently, more than 100 models of converters are availablein the market. The existence of these conversion technologiesmay enable compatibility across different formats withoutcompromising product performance or features (Farrell et al.1992). This aspect of the flash memory market has nottypically been observed in other markets for analog productswith network effects. For example, converting between theVHS and Betamax videotape standards was costly andresulted in lower quality output (Cusumano et al. 1992).

These observations suggest a second research question: Howdo digital converters influence standards competition in theflash memory card market? We theorize that, in contrast toother IT markets studied in the network effects literature (e.g.,Song and Walden 2003, 2009), a candidate explanation forthe apparent anomalies in the competitive dynamics of flashmemory cards involves both the digital nature of flashmemory products and the associated presence of conversiontechnologies. Ultimately, however, the direction and size ofthe impact of flash memory converters on competitivedynamics in this market is an empirical question.

In this paper, we attempt to test the impact of flash memoryconverters directly by analyzing a unique and extensive dataset primarily obtained from the market research firm NPD.The data include monthly retail prices and unit sales for flashmemory converters and for all six flash memory card formatssold from 2003 to 2006. We first use this data to test whetherthere is a product price premium for formats with largerinstalled bases—a standard test in the literature for thepresence of network effects (Brynjolfsson and Kemerer 1996;Gallaugher and Wang 2002; Gandal 1994).

Finding the presence of network effects as measured in thismanner, we then analyze why these network effects may nothave led to rapid market dominance, as seen in other settings.Specifically, we hypothesize and find that the increasingadoption of digital converters reduces the impact of theinstalled base on the product price premium. In particular,

digital converters reduce the price premium of the leadingformats more than they do that of the minority formats. Notonly is the impact of the installed base less significant whenthere is greater adoption of digital converters, but marketconcentration also decreases as digital converters are widelyadopted, suggesting increasing competitiveness among flashmemory card manufacturers with converter introduction. Thus, our results suggest that, in the presence of low costconversion technologies and digital content, the probability ofmarket dominance can be lessened to the point where multipleotherwise incompatible standards are viable.

In addition to addressing a relevant and practical questionregarding potential winner-take-all markets, our findings haveboth theoretical and managerial implications for the growingliterature on standards competition in digital goods markets. Specifically, our empirical study complements the analyticliterature on conversion technologies in markets withstandards competition (Choi 1996, 1997; Economides 1989,1991; Farrell and Saloner 1992; Liu et al. 2011; Matutes andRegibeau 1988). Although the presence and magnitude ofnetwork effects have been empirically demonstrated in theliterature (Asvanund et al. 2004; Brynjolfsson and Kemerer1996; Gallaugher and Wang 2002; Gandal 1994), our currentempirical work suggests differences in the nature of networkeffects in digital goods technologies, and documents anassociated interaction between network effects, converters,and market evolution. Our empirical analysis thus contributesby providing new insights into how conversion technologiesaffect pricing strategies in digital markets, which are of signi-ficant and increasing practical commercial importance.

Our findings on the effects of conversion technologies alsohave potentially important implications for both vendors andconsumers. As converters become more popular, consumerperceptions of the value of network effects decrease, sincecompatibility can be achieved at a lower cost. Consequently,the choice of a product may rely more on factors other thanmarket share, such as brand and quality attributes. As con-sumers broaden their choice sets, vendors’ marketing andpricing strategies should adjust accordingly. Thus, the con-sumer decision-making process and the interaction betweenvendors and consumers may change significantly as a resultof the introduction of converters.

Finally, from society’s standpoint, the provision of a con-verter reduces the need to compromise between productvariety and standardization, especially for markets character-ized by high consumer heterogeneity. Given that it is difficultto achieve industry-wide compatibility without lengthy andcostly coordination, our analysis provides an alternative wayto overcome the compatibility barrier without incurringsignificant costs of standardization.

11In our study a converter is a device that can accept at least two flashmemory card formats.



The paper proceeds as follows: the next section brieflyreviews the literature on network effects and conversiontechnologies, providing the theoretical basis for our work. Thesubsequent section presents the conceptual model andhypotheses. We then describe the data and measures of ourkey variables. In the following sections, the econometricmodels and results are presented and further discussed.Finally, we present our conclusions and suggest directions forfuture research.

Related Literature

Network Effects and Hedonic Price Models

Network effects refer to circumstances in which the net valueof consuming a good (e.g., subscribing to telephone service)is affected by the number of agents taking equivalent actions(Katz and Shapiro 1985). Prior research has suggested that,in product markets with network effects, early success inaccumulating a large installed base of customers can give riseto a number of strategic advantages. In addition to the posi-tive feedback loop generated by self-fulfilling consumer andretailer expectations, network effects help to create switchingcosts and lock-in among existing customers (Chen and Hitt2002; Zhu et al. 2006) and to increase the speed at whichmarket demand grows (Economides and Himmelberg 1995;Kauffman et al. 2000). Other strategic advantages of networkeffects include the ability to deter potential entrants (Lee et al.2003; Suárez and Utterback 1995) and the possibility tocontrol the design interface (Conner 1995). Moreover, asnetwork effects are often perceived to follow the consumer’svaluation for a standard (Farrell and Saloner 1985), a streamof empirical research on network effects focuses on esti-mating the influence of the installed base on a consumer’swillingness to pay for the dominant standard.12 Severalempirical studies have found a significant advantage fordominant standards in markets for IBM compatible micro-computers (Hartman 1989), mainframe computers (Greenstein1993), spreadsheet software (Brynjolfsson and Kemerer 1996;Gandal 1994), databases (Gandal 1995). and communicationsequipment (Chen and Forman 2006).

Although these empirical studies differ in their highlightedantecedents of network effects, they all adopt the same notionof a price premium in interpreting the estimated coefficientson the compatibility variables. A price premium refers to theprice advantage a product enjoys over the other competingproducts due to particular distinct product features such asbrand, quality, or, in the above studies, compatibility. A num-ber of techniques have been used to identify the pricepremium resulting from network effects; of particular interestis the use of hedonic regressions in capturing this value.

Hedonic regressions were first applied to IT products byChow (1967) in estimating the annual quality-adjusted pricedecline in mainframe computers from 1960 to 1965. As auseful method to disaggregate consumers’ consumption utilityinto independent valuations of different aspects of a product,the hedonic regression has been widely employed in esti-mating the marginal benefit of products that include multipleattributes, and has been usefully employed in the empiricalliterature on network effects. The hedonic approach regressesa product’s listed price on a number of product attributes, andthe coefficients on these attributes represent the price premiaassociated with these attributes. By treating various ante-cedents of network effects, such as the size of the installedbase or learning costs, as implicit features of a product,hedonic regressions allow researchers to obtain estimates ofthe parameters capturing a consumer’s willingness to adopt astandard (or the opportunity costs to switch to a differentnetwork). Hence, the price premium can be computed as theportion of the listed price that is attributed to the size of theproduct installed base, controlling for other intrinsic values ofa product. As the choice of a flash memory card also dependson a variety of considerations other than the size of the in-stalled base (e.g., brand, capacity, speed), we follow theliterature and adopt the hedonic framework in our study as anappropriate approach to distinguish the impact of networkeffects from other factors.

Conversion Technologies

An important objective of this research is to analyze the roleof digital converters in influencing standards competition inmarkets with network effects. Although the extant empiricalliterature has identified a variety of sources and consequencesof network effects, little attention has been devoted to theinteraction between conversion technologies and technologyadoption in markets characterized by standards competition.The studies that do address this topic have relied on eitheranalytic frameworks (Choi 1996, 1997; Farrell and Saloner1992), or an historical case study (David and Bunn 1988) toillustrate the effect of converters on technology adoption.

12There is also a stream of research that uses elasticities to measure indirectnetwork effects (i.e., Clements and Ohashi 2005; Nair et al. 2004). However,the flash memory card market differs from the markets studied in these papersas it exhibits both direct and indirect network effects. Since, as a practicalmatter, it would be extremely difficult to collect sales data on all possibledigital devices that use a flash memory card (to measure indirect networkeffects), we follow the lead of other researchers and adopt a direct effects(hedonic price) model in our paper.



There are no large-scale empirical studies of this phenomenonof which we are aware.

The lack of empirical studies on this topic may derive fromthe difficulty in distinguishing the counteracting effects ofconversion technologies on product price. In the absence ofa common interface, converters enable incompatible systemsto communicate with each other and hence internalize thecompatibility benefits that would have been lost without con-verters. Prior research has suggested that consumer benefitsvia the provision of converters include both greater productvariety and the increased size of the network to which theconsumer belongs (Economides 1989; Matutes and Regibeau1988). As a result, consumers are willing to pay a higherprice for otherwise incompatible products. On the other hand,the presence of converters also reduces the expected pricepremium of the dominant standards as both the relative attrac-tiveness of their products and product switching costsdecrease due to a lower compatibility barrier (Farrell andSaloner 1992). The installed base of the dominant standardsmay expand more slowly and competition may intensify asthe intransient incompatibility period extends (Choi 1996,1997). At the same time, new entrants are more likely toenter the market and to survive the standards competition (Liuet al. 2011).

Given these complicated interactions, the product pricepremium is not merely an indicator of the perceived value ofthe installed base as it is typically modeled in the networkeffects literature. A consumer’s valuation of product compati-bility, as measured by product price, needs to be furtherdisaggregated into variation due to the product’s installedbase, the adoption of conversion technologies, and the inter-action of the two effects. Drawing on the findings from theabove literature, we develop a conceptual model with specifichypotheses to examine the dynamics between conversiontechnologies and the various antecedents and outcomes ofnetwork effects.

Research Model and Hypotheses

Since the role of conversion technologies is of primaryinterest when network effects exist, identifying the presenceand magnitude of network effects is an important first step inevaluating the nature of standards competition in the flashmemory card market. Following the literature, we adopt ahedonic model to measure network effects in the flashmemory market, where the price premium of the installedbase measures the value a consumer places on the size of aflash memory format’s installed base (and thus networkeffects), controlling for other flash memory card attributes

such as brand, capacity, and speed. Note that since the pricepremium only measures the relative impact of a particularproduct attribute (i.e., installed base) on listed price, it doesnot always move in the same direction as listed price. Thisdistinction is important in the IT industry—and in the flashmemory card market in particular—where a declining productprice is commonly observed due to rapid technologicaldevelopment. Although the listed price is declining, a productformat could still enjoy a positive price premium from itsinstalled base (a positive coefficient of the installed basevariable in the hedonic regression) if consumers believe thatcompatibility is important.

Our first hypothesis considers the effects of market power onprice premiums for leading vendors. In particular, the size ofthe installed base may play a dual role with respect to theprice premium of a flash memory card. On one hand, whennetwork effects are present, a larger installed base for aproduct format confers greater utility to consumers. Hencethe price premium of a flash memory format could varypositively with the size of the flash memory format’s installedbase. For example, in the context of software, due to indirectnetwork effects, consumers of the dominant operating systemwill find that they can use more software applications thanusers of a competing, but minority, system. This is similar tothe case where owners of a dominant flash memory cardformat will find that their flash memory cards are supportedby more digital devices than owners of a less popular flashmemory card.

On the other hand, due to economies of scale in production,a flash memory format with a larger sales volume can enjoya greater cost advantage over those with a smaller salesvolume. Hence the price premium of a flash memory formatcould also vary negatively with the size of its installed base.

According to the classic network effects theory, if the utilityof a product increases with the installed base for the product,there will tend to be one dominant standard in the market, andthe firm offering this standard should be able to charge ahigher price, reflecting the higher value that consumers per-ceive. Given the magnitude of network effects, we expectthat if there exists a strong demand for compatibility, the pricepremium due to market power will dominate the price reduc-tion effect due to production economies of scale, such that

Hypothesis 1: The price of a flash memory card ispositively associated with the size of the installedbase of the same format.

Our second hypothesis considers how the introduction ofdigital converters affects the price of flash memory cards,arguing that digital converters increase the usefulness and



thereby the value of flash memory cards in general (i.e., at anaverage level of the installed base). When digital convertersare available, a consumer who owns a flash memory card ofa particular format can exchange data not only with the otherdigital devices she owns and other devices within the sameflash memory network, but also with out-of-network con-sumers who own digital devices that use incompatible flashmemory cards, and can thereby obtain the benefits of com-patibility. Figure 2 illustrates the expanded consumer choiceset that is afforded by converters. Thus digital converters canincrease the consumption utility across flash memory cardproducts of different standards. This implies that greateradoption of digital converters will increase the overall utilityfor flash memory cards, even if the cards are not compatible. Therefore,

Hypothesis 2: The prices of flash memory cards arepositively associated with the adoption of digitalconverters.

Our third hypothesis considers the interaction betweenproducers’ market power and the introduction of digitalconverters, and posits that the presence of convertersincreases product substitutability and thereby reduces thevalue of flash memory cards, especially for the dominantproducers (i.e., those with an installed base that is larger thanaverage). When making a technology choice, the widepresence of digital converters reduces the consumer’s risk ofbeing stranded on a new, but less popular standard, as thechances for survival of a new technology are larger whennetwork effects are less significant. In addition, digital con-verters allow consumers who own incompatible products toexchange data with each other. As a result, when digital con-verters are widely present consumers are not as motivated topurchase a dominant standard as there is less benefit from it;this lowers the producer’s market power and, consequently,its price premium due to network effects. Following this logicin the context of flash memory cards, a greater adoption ofdigital converters will especially affect the price of thedominant standard. Producers of flash memory card standardswith a larger installed base are expected to lose more marketpower than those with a smaller installed base, as they havemore value to lose from being a dominant standard whenconverters are present. Thus, we expect that

Hypothesis 3: The adoption of digital convertersreduces the impact of the installed base on flashmemory card prices such that the price reductioneffect is stronger for products with a larger installedbase than for products with a smaller installed base.

Finally, in Hypothesis 4, we consider the effects of digitalconverters on market concentration for flash memory card

producers. Classic network effects theory predicts that pro-duct markets will tip toward a single dominant standard whenthere are strong network effects. Consequently, market con-centration will typically increase once the installed base of theleading standard has reached a critical mass. However, asargued here, it is possible that the presence of conversiontechnology will affect the nature of competition, as con-version technology can offset some of the impact of networkeffects. If this is true, it is less likely that a dominant pro-ducer will emerge in a market with an increasing presence ofconverters. The flash memory card market could then beexpected to be less likely to tip toward one dominant produceras many different formats can be converted to becomecompatible. Therefore,

Hypothesis 4: Market concentration of flashmemory card producers decreases as the adoptionof digital converters increases.

Figure 3 summarizes these four hypotheses and illustrates theconceptual framework for our empirical analysis along withthe predicted directions of the hypothesized interactionbetween the adoption of digital converters and a product’sinstalled base. This figure also illustrates two importantmarket outcomes: card price and market concentration. Con-trol variables are shown in dashed boxes.

Data and Measures

Sample

To test our hypotheses, we assembled a large panel data setincluding data on flash memory card products and theirproducers. We selected a sample period from 2003 to 2006for our analysis as this is a critical period in the developmentof the flash memory card market during which all six majorformats are present. Our primary data were generouslyprovided by the NPD research group. These data includedetailed information on monthly retail prices and unit salesdata of the major flash memory cards and digital converterssold each month by major U.S. retailers. These data areobtained by NPD directly from point-of-sale (POS) terminalsin major retailers across a range of outlets and cover theperiod January 2003 to August 2006.

To supplement the NPD data set, we also implemented asoftware agent to retrieve daily observations of flash memorycard prices, sales rank, and product review data fromAmazon.com. We use the customer review ratings to controlfor the reputation of different flash memory card models, andwe use the price data to validate the retail prices from theNPD data set. Finally, we gathered the flash memory cards’



Figure 3. Research Model and Hypotheses

product specification data from each flash card format’sofficial trade association.

The final data set consists of 15,091 observations of 706 pro-duct panels13 across 44 months, and covers all six major flashcard formats and 45 major brands, with capacities rangingfrom 4 megabytes to 8 gigabytes. Each product panel repre-sents a format i, brand j, capacity k flash memory card soldduring month t. The product level panels allow us to controlfor variation due to formats, capacities, and brands, whereasthe time series data allow us to control for variation due topotential “seasonal fluctuations” (e.g., holiday sales surge)and time trends (e.g., declining costs). The distribution ofobservations, broken down by format and year, is shown inTable 1.

Variables

Table 2 provides definitions of the key variables used in ouranalysis.

The key variables for our analysis are card price and theinstalled base of a flash memory card model, as well as theadoption level of converters. We compute flash memory cardprice, CardPricei,j,k,t, as the deflated (in 2003 Q1 dollars)average retail price of flash memory cards of format i, brandj, and capacity k sold during month t (Brynjolfsson and

Kemerer 1996). The current installed base, InstalledBasei,t, iscomputed as the cumulative units of format i compatible flashmemory cards sold up to month t. The level of digital con-verter adoption, ConverterAdoptiont, is the cumulative num-ber of the digital converters sold up to month t. Our controlvariables include the product’s capacity and speed, to accountfor possible variations due to memory card capacity andproduct specifications, as well as six format dummy variablesto capture other format specific product features. Moreover,since different flash memory cards may be at different stagesof their own product life cycles and different products mayexperience different levels of competition within their ownproduct category, we also construct two variables—productlife cycle and intra-format competition at the product level tocontrol for these effects.14 The variable product life cycle isa binary variable and is coded as zero for flash memory cardsthat were discontinued before the end of our sample periodand one for flash memory cards that were still being sold onthe market at the end of our sample period. The variableintra-format competition is measured by counting the numberof brands in a given product capacity category. To control forprice premium due to product reviews, we create a reputationvariable using the average of the Amazon customer reviewratings for the product.15 Finally, we create three otherdummy variables to control for brand, seasonal and yeareffects.

13Note that this is not a balanced panel due to the fact that some brands mayonly produce a type of flash memory card at certain capacities, and someflash memory cards are discontinued after a certain period of time.

14We thank two anonymous referees for their suggestions to include thesetwo control variables.

15We also used Amazon’s sales rank as an alternative measure of this vari-able. However, both measures turn out to be insignificant in our estimation.

Network Effects(Installed Base)

Capacity, Speed, Format, Brand, etc.

Converter Adoption

Card Price

MarketConcentration

Price Variation

H1 (+)

H4 (-)

H2 (+)H3 (-)



Table 1. Distribution of Flash Memory Card Observations and Annual Sales by Card Type/Year*

Card Type 2003 2004 2005 2006** Total (%)

Compact Flash 1,205 1,309 1,412 921 4,847 (32.12%)

3,504,914 3,955,887 3,050,276 1,404,866 11,915,943 (17.65%)

Memory Stick 334 453 492 343 1622 (10.75%)

2,745,576 3,460,319 4,452,578 3,270,090 13,928,563 (20.63%)

Multimedia 317 360 453 267 1,397 (9.26%)

858,478 1,905,180 775,550 337,450 3,876,658 (5.74%)

Secure Digital 742 1,063 1,386 1,251 4,442 (29.43%)

3,531,643 7,309,169 10,027,432 7,110,514 27,978,758 (41.43%)

Smart Media 583 461 387 200 1,631 (10.81%)

1,044,834 461,388 261,202 76,019 1,843,443 (2.27%)

xD Picture 190 264 385 313 1152 (7.63%)

1,306,797 2,195,741 2,864,649 1,608,493 7,975,680 (11.81%)

Grand Total 3,371 3,910 4,515 3,295 15,091

12,992,242 19,287,684 21,431,687 13,807,432 67,519,045

*The first row in each flash memory category shows the number of observations and the second row shows annual unit sales.**Note that 2006 observations include up to August, 2006

Table 2. Definitions of Key Variables

Variable Name Definition

CardPriceDeflated (in 2003 Q1 dollars) average retail price of a format i, brand j and capacity k flashmemory card sold during month t.

InstalledBase Cumulative number of format i flash memory cards sold up to (and including) month t.

ConverterAdoption Cumulative number of digital converters† sold up to month t.

Capacity Capacity (in MegaBytes) of a flash memory card.

Speed Average read/write speed of a flash memory card.

Product Life CycleA dummy variable that measures a flash card’s life cycle stage, coded as zero for products thatwere discontinued before the end of our sample period and one for products that were still beingsold on the market by the end of our sample period.

Intra-formatCompetition

The number of brands within a given flash memory card category.

Reputation The average customer review rating on amazon.com for the flash memory card of format i, brandj and capacity k, ranging from 1 (low) to 5 (high).

Compact Flash Dummy variable, 1 if the flash memory card is compatible with the Compact Flash format.

Memory Stick Dummy variable, 1 if the flash memory card is compatible with the Memory Stick format.

Multimedia Dummy variable, 1 if the flash memory card is compatible with the Multimedia Card format.

Secure Digital Dummy variable, 1 if the flash memory card is compatible with the Secure Digital format.

Smart Media Dummy variable, 1 if the flash memory card is compatible with the SmartMedia format.

xD Picture Dummy variable, 1 if the flash memory card is compatible with the xD Picture format.

D_Brand A “make effect” dummy variable, 1 if the flash memory card is manufactured by Firm j*.

D_Quarter A seasonal effect dummy, 1 if the observation belongs to quarter q (q = 1, 2, 3 or 4).

D_Year A year dummy, 1 if the observation belongs to year y (y = 2003, 2004, 2005 or 2006).

*Only brands with market share greater than 1% are selected.†We only include converters that can read from and write to multiple flash memory formats, as they represent the majority of the flash memory cardssold in the market (more than 90% based on our data) and they better capture the demand for compatibility than single-format converters whichonly transfer data between flash memory cards and PCs.



Table 3. Descriptive Statistics of Key Variables

Variable Mean Std. Dev.

CardPrice 61.08 85.29

InstalledBase 18951540 7983519

ConverterAdoption 234874 162905

Capacity 420.74 804.97

Speed 20.68 15.04

Product Life Cycle (Dummy) 0.793 0.405

Intra-format Competition 3.71 1.53

Reputation (review rating) 3.12 1.69

Table 4. Correlations of Key Variables†

CardPrice

InstalledBase

ConverterAdoption Capacity Speed Reputation

ProductLife Cycle

Intra-format

Competition

CardPrice 1.000

InstalledBase 0.049** 1.000

ConverterAdoption -0.099** 0.069 1.000

Capacity 0.637** 0.183** 0.276** 1.000

Speed 0.192** 0.137* 0.016 0.231** 1.000

Reputation 0.076* 0.183 -0.044 -0.149* 0.357** 1.000

Product Life Cycle 0.165** 0.246** 0.253** 0.187* 0.059 0.182 1.000

Intra-formatCompetition

-0.131** 0.267** 0.226** 0.215 0.043 0.325* 0.150 1.000

*p < 5%**p < 1%†As converters convert between multiple formats, at any given point of time they could have an impact on prices of flash memory cards of all formatsat all capacities. Therefore, within a given month, we can replicate the value of ConverterAdoption for each observation of flash memory card andperform the correlation test. In other words, when computing the correlation between the CardPrice variable and the ConverterAdoption variable,the CardPrice column consists of a number of data points with different values and the ConverterAdoption column consists of the same numberof data points with an identical value.

We present descriptive statistics in Table 3 and the correla-tions of the key variables in Table 4. The intercorrelationsbetween the variables in our model are generally low. Asexpected, two control variables, capacity and speed, are bothpositively correlated with flash memory card price, with cardshaving larger capacities and faster speeds commanding higherprices, ceteris paribus. The other two key variables, flashcard installed base and converter adoption, are modestlycorrelated with price, with installed base having a positivecorrelation and converter adoption a negative correlation.

Econometric Models, Estimation,and Results

Network Effects, Digital Converters,and Price Premia

We construct several econometric models to test our hypoth-eses. Given that the proposed effect of converters is non-linear, it should not be modeled as a linear predictor of pro-



duct price as has been done in the classic linear hedonicmodels. Therefore, to test Hypotheses 1 through 3 we con-struct an interaction model as shown in Model 1.16

CardPricei,j,k,t = α0 + α1InstalledBasei,t–1 +α2ConverterAdoptiont–1 + α3InstalledBasei,t–1 *ConverterAdoptiont–1 + α4Capacityk + α5Speedi,j,k +α6Reputationi,j,k,t + α7ProductLifecycle_Dummyi,j,k,t

+ α 8 I n t r a F o r m a t C o m p e t i t i o n i , k , t +α9Format_Dummy i+ α10Brand_Dummy j +α11Year_Dummyt + α12Quarter_Dummyt + gi,j,k,t

[Model 1]

When the variable ConverterAdoption and its associatedinteraction term are both absent, Model 1 reduces to a classichedonic price regression where the coefficient α1 representsthe impact of the installed base on product price and isexpected to be positive and significant when there are strongnetwork effects. However, when these two variables areincluded in the model, the marginal effect of the installed baseon product price is not solely captured by coefficient α1.Instead, the impact of the interacting variable also needs to betaken into account. More specifically, the marginal effect onproduct price should be computed as the partial derivative ofthe dependent variable with respect to the variable of interest. Thus, Hypothesis 1, which predicts that, ceteris paribus, alarger installed base will increase the price of a flash memorycard, can be represented as

H1: = α1 + α3ConverterAdoptiont–1 > 0,∂

∂Card

InstalledBasei j k t

i t

Price , , ,

, −1

when evaluated at the mean of ConverterAdoptiont–1.

Similarly, Hypothesis 2, which predicts that the adoption ofdigital converters will lead to a higher flash card price, can berepresented as17

H1: = α2 + α3InstalledBasei,t–1 > 0,∂

∂Card

ConverterAdoptioni j k t

t

Price , , ,

−1

when evaluated at the mean of InstalledBasei,t–1.

Finally, Hypothesis 3, which focuses on the interactionbetween the installed base and converters, can be tested byexamining the significance level of coefficient α3, and by

conducting an F-test on the restricted model (the one withoutthe interaction term) versus the unrestricted model (the onewith the interaction term).

We first estimate a restricted model and then include theinteraction term in an unrestricted model to examine if thecoefficient estimates and model fit statistics are sensitive tothis specification change. Other variables in the restrictedmodel include capacity, speed, product reputation, productlife cycle, and intra-format competition, as well as format,brand, and seasonal and yearly dummies. The brand dummiescover the top 10 flash card brands in our data set. Theomitted (base) dummy variables for the other categories arethe SmartMedia format, winter quarter, and the year 2006.Therefore, the constant term estimated in the model may beinterpreted as the predicted price of a non-major brandSmartMedia card sold in the last quarter of 2006.

The ordinary least square (OLS) regression results of both therestricted and unrestricted models are provided in Table 5. Inthe restricted model, the coefficient for InstalledBasei,t–1 ispositive, and the coefficient for ConverterAdoptiont–1 isnegative, and only the latter is significant at the 1% level.When the interaction term is included in the restricted model,the signs of the coefficients for both InstalledBasei,t–1 andConverterAdoptiont–1 remain unchanged, but both becomesignificant at the 1% level. The interaction term is negativeand significant at the 1% level as well. The signs of the coef-ficient estimates of the control variables are largely consistentwith our expectation: the coefficients for capacity, speed, andproduct life cycle are all positive and significant, suggestingthat a larger capacity, higher speed, and current generationflash memory cards are associated with a price premium. Thecoefficient for intra-format competition is negative andsignificant, indicating that more competitors in the sameproduct category will lead to a more intense competition andhence a lower average flash memory card price.18 The coeffi-cient for reputation is positive, but only marginally significantin the unrestricted model. Finally, the decreasing coefficientestimates of the yearly dummies show that flash card pricedeclines over time. The results are summarized in Table 5.

16Note that although quarterly dummy variables are used in this equation, theresults are equivalent to a specification with monthly dummy variables.

17Note that in the expressions to evaluate H1 and H2, α3 is the same coeffi-cient obtained from the estimation of Model 1.

18We also evaluated an alternative measure of this variable using the standarddeviation of prices from different vendors selling the same product (a smallervalue indicates that there is more intense competition within this productcategory). This alternative measure yields a consistent result.



Table 5. Regression Results†‡ – Model 1

Dependent Variable: CardPricei,j,k,t

OLS RegressionRestricted (Nointeraction)

OLS RegressionUnrestricted(Interaction)

GLSEstimation

(Interaction) 2SLS Estimation

(Interaction)

Constant15.13

(4.29)**30.89

(4.65)**6.38

(3.89)**16.78

(2.91)**

InstalledBasei,t–1

3.57e-07(1.45e-07)*

1.74e-06(2.60e-07)**

1.93e-06(1.30e-07)**

5.21e-06(2.48e-07)**

ConverterAdoptiont–1

-.00015(1.51e-07)**

-.0001845(.0000156)**

-.0000689(4.17e-06)**

-.0000922(.0000191)**

InstalledBase * ConverterAdoptiont–1

-8.65e-12(8.93e-13)**

-1.44e-11(3.69e-13)**

-1.04e-11(7.67e-13)**

Capacity.0772

(.00070)**.0773

(.00069)**.068

(.0011)**.054

(.0017)**

Speed 1.56(.19)**

1.52(.186)**

0.62(.178)**

1.57(.648)**

Reputation 4.32(2.17)*

3.55(2.41)

2.15(1.77)

1.66(1.12)

Dummy Product Life cycle18.14

(2.97)**17.85

(1.35)**12.17

(1.16)**33.01

(4.35)**

Intra-format Competition-1.14

(0.27)**-1.76

(0.13)**-0.63

(0.047)**-2.74

(0.146)**

Dummy_2003 27.26(11.93)*

58.29(12.32)**

3.87(1.68)*

24.75(5.56)*

Dummy_2004 23.11(9.38)**

43.36(9.49)**

2.34(1.18)*

12.68(4.32)*

Dummy_2005 14.48(5.13)**

16.78(5.12)**

1.06(.49)*

3.53(2.46)*

Adjusted R²0.5609 0.564 Log likelihood =

–436710.5014

Fit StatisticF(25, 15065) =

731.51**F(26, 15064) =

768.39**Wald P2 (26) =

6045.61**Wald P2 (26) =

8139.60**

†N = 15,091, Ni = 6, Nj = 45, Nk = 44 (706 panels across 44 months). ‡Standard errors in parentheses. *p < 5%. **p < 1%.

Before we proceed to interpret the coefficients and test thehypotheses, we also conduct several robustness analyses19 toensure that our results are robust to various specificationerrors and violations of the OLS estimation assumptions. The multiplicative nature of our model implies that themarginal effects of the linear variables can be confoundedwith the influence of the interaction term. To make our

interpretation of the linear terms more straightforward, wefollow the current practice in the literature (Aiken and West1991, pp. 35-36; Jaccard et al. 1990) and center the originalinteracting variables before computing the interaction term. This is done by subtracting the mean from every observationfor both interacting variables. After centering, the means ofthe centered variables are zero, and the correlations betweenthe interaction term and the original variables are muchsmaller.20 A multicollinearity check also reveals that, aftercentering, the condition number of the interaction model andthe variance inflation factor (VIF) values of the interactionterm and the original variables are both below the recom-mended threshold values of 20 and 10, respectively (Greene

19As some of the literature on network effects considers a nonlinear speci-fication, we also tested a model with an additional variable, a squared termof the installed base, to examine whether the impact of converters issensitive to the specification of network effects. The results indicate thatnetwork effects increase at a faster rate as the size of the installed baseincreases (the coefficient estimate for the squared term is positive andsignificant at the 1% level), but the effects of converters on flash card pricesare still qualitatively consistent with those obtained from the linearspecification. Therefore, for ease of interpretation, we report the resultsobtained from the linear specification of the network effects model.

20For ease of comparison, both InstalledBasei,t–1 and ConverterAdoptiont–1

are centered in the OLS regressions.



2003, pp. 57-58). Therefore, we conclude that our interactionmodel does not exhibit excessive multicollinearity.

A Breusch–Pagan/Cook–Weisberg test for heteroskedasticityyields a value of χ²(1) = 5120.7 (p < 0.001), suggesting thepresence of heteroskedasticity. This is consistent with a plotof the residuals versus fitted (predicted) values, whichexhibits a wider scatter with greater X-axis values.

The Wooldridge test for autocorrelation in the panel datashows that first-order autocorrelation (AR1) cannot be ruledout for our data set. This is not surprising, given the longi-tudinal nature of our data. The presence of heteroskedas-ticity and AR1 autocorrelation would argue against the useof OLS (Greene 2003). As both heteroskedasticity and auto-correlation are present in our data set, we address bothproblems simultaneously by applying a generalized leastsquares (GLS) estimation procedure21 with corrections toadjust for both heteroskedasticity and panel-specific first-order autocorrelation. As shown in Table 5, column 3, ourresults are robust to these corrections. The interaction effectremains significant and the directions of the estimated coeffi-cients for both the interacting variables and the interactionterm remain the same in the GLS regression, although theprice increase effects brought by converters to minorityformats are smaller compared to those obtained from theOLS regressions.

As the GLS estimator is more robust to heteroskedasticityand autocorrelation, we examine both marginal effects andinteraction effects using the results from the GLS estimation.To evaluate Hypotheses 1 and 2 in Table 6, we compute themarginal effects of the variables InstalledBase and ConverterAdoption at the means of the interacting variables.22 Fol-lowing Greene (2003, p. 124), the standard errors of thesemarginal effects can be computed from

and similarly from

Note that the standard errors of the marginal effects at dif-ferent values can be obtained by substituting the respectivevariables with these different values into the above equa-tions. These standard errors are provided in parentheses.

The “Mean” column of the first row in Table 6 shows thatthe marginal effect of the flash memory card installed baseis positive at the mean value of the converter adoption level.This provides support for Hypothesis 1, suggesting thatnetwork effects as measured here do exist in the flashmemory card market such that the price of a flash memorycard is positively associated with the size of the installedbase for the same format. The “Mean” column of the secondrow shows that the marginal effect of converter adoption isnegative when evaluated at the mean of the installed base. Thus, the price of a flash memory card is negatively asso-ciated with the level of converter adoption, and Hypothesis 2is not supported. With regard to Hypothesis 3, as shown in Table 6,the flashcard price premium changes in the expected direction. Thesecond row shows that the marginal effect of digital con-verter adoption is larger (in absolute value) for flash memoryformats with a larger installed base (the +1 and +2 standarddeviation columns) than those with a smaller installed base(the –1 and –2 standard deviation columns). Moreover,coefficient α3 is highly significant (p < 0.001) in the inter-action model. An F-test of the difference between therestricted model and the unrestricted model confirms that theinteraction term explains variance in the hedonic regression. Therefore, Hypothesis 3 is supported.

One possible concern about these results is that one of theindependent variables, InstalledBase, is the cumulative unitsales volume of a flash memory card format. This variablecould be closely correlated with the current period unit salesvolume of the same flash memory card format, which, inturn, could be correlated with our dependent variable, thecurrent period flash memory card price. To address thispotential endogeneity, we perform a two-stage least square(2SLS) estimation in the following two steps: (1) we first-difference the CardPrice variable and the InstalledBasevariable to compute their residual values, and (2) we use thelagged term of the differenced InstalledBase variable as ourinstrumental variable and the differenced CardPrice variableas our dependent variable and perform a 2SLS estimation. Since the lagged difference of the InstalledBase (the vari-able) is uncorrelated with the present difference of CardPrice

21We also estimated a fixed effects model. Our main results still hold.However, a fixed effect specification results in excessive collinearity (manyof the control variables are dropped in a fixed effect model). Therefore, weadopt the GLS estimator.

22We also compute the marginal effects at different values of the interactingvariables to show how the marginal effects change as the value of theinteracting variables varies.

[ ]

[ ] [ ]

VarE CardPrice InstalledBase ConverterAdoption

InstalledBase

Var InstalledBase Var InstalledBase Cov

i j k t i t t

i t

i t i t

∂∂

α α α α

, , , ,

,

, ,

| ,

( ) [ ) ,

=

+ +12

3 1 32

[ ]Var

E Cardprice InstalledBase ConverterAdoption

ConverterAdoption

i j k t i t t

t

∂∂

, , , ,| ,



Table 6. Marginal Effects

-2 Std. Dev. -1 Std. Dev. Mean +1 Std. Dev. +2 Std. Dev.

Marginal effect ofInstalledBasei,t–1

6.62E-06(3.42E-06)*

4.28E-06(0.62E-06)**

1.93E-06(0.26E-06)**

-4.16E-07(0.66E-07)**

-6.62E-06(1.14 E-06)**

Marginal effect ofConverterAdoptiont–1

1.61E-04(0.86 E-04)*

4.61E-05(0.83 E-05)**

-6.9E-05(0.94E-05)**

-1.8E-04(0.43E-04)**

-2.99E-04(0.52E-04)**

*p < 5%

**p < 1%

(the dependent variable), and is highly correlated with thepresent difference of the InstalledBase variable, it satisfies therequirements for an instrument. Following Baum, Schaffer,and Stillman (2003) we tested for endogeneity in ouraugmented form using a generalized methods of moments(GMM) estimation with specifications for autocorrelation andheteroskedasticity.23 The results from the GMM estimation(summarized in Table 5) are consistent with those obtainedfrom the OLS and GLS estimations. Neither a Wu–HausmanF test (p-value = 0.31) nor a Durbin–Wu– Hausman chi-square test (p-value = 0.35) could reject the null hypothesisthat the lagged cumulative market share is exogenous. Thisprovides confidence in our results against potentialendogeneity concerns.

In addition to these corrections, we performed several sensi-tivity analyses to ensure the robustness of our results. Onepotential issue is the computation of the InstalledBasevariable in the initial sample period. Since some formats hadbeen in the market for several years before our sample period,the installed base of these formats should also include theirprior sales figures. Although NPD did not have flash memorycard sales data prior to 2003, we were able to obtain addi-tional annual sales data of these six flash memory formatsprior to 2003 from another independent market research firm. We found that including pre-2003 cumulative sales data doesnot materially change our results.24

Since our data set is an unbalanced panel, we also examinewhether selection bias exists in our sample. Our results will

be biased if there is a relationship between our dependentvariable and some inherent characteristics of the missing data. Following Verbeek (2004) and Verbeek and Nijman (1996),we took the following steps to check for potential selectionbias. First, we removed the observations that do not exist forthe entire period of our sample and created a balanced sub-panel. Second, we estimated both fixed and random effectsmodels on the balanced subpanel and contrasted the resultswith those obtained from the original unbalanced set. Wefound that the vector of coefficients and the variance–covariance matrix for the balanced panel and for theunbalanced panel are not significantly different from eachother, suggesting that there is no significant selection bias inour estimation.

Finally, the data on converters used in the paper pertains onlyto external multiformat converters. We note that many com-puters and printers are sold with integrated multiformatreaders, which are not included in the study. To address thisissue, we performed a sensitivity analysis on the measure ofthe ConverterAdoption variable used in our analysis. Wefound that in order for our results to become insignificant, themeasure of the ConverterAdoption variable needs to be 12.87times larger than the current measure of the ConverterAdoption variable.25 Since it is not reasonable to expect thatthere are nearly 13 times as many converters bundled withPCs and printers as those sold directly in the market, webelieve that our results are robust to this data limitation.

The Effect of Converters onMarket Concentration

When converters are not available in markets with strongnetwork effects, a large installed base can give firms a signi-

23The same approach is also used in Mittal and Nault (2009) in a similarsituation.

24We believe that this is due to two reasons. First, the inclusion of new datamakes the impact of the converters’ price reduction for leading formats moresignificant (as the installed base measures for those initially leading formatsare even larger than before). Second, the impact of converters is relativelysmaller in 2003 than later in our sample period, and the sales volume prior to2003 is relatively smaller than that after 2003, hence it does not lead to asignificant change in our results.

25To perform this sensitivity test we multiplied our ConverterAdoptionmeasure step by step (i.e., X2, X3, X4,...X13) until the results becomeinsignificant, and then calculated backward to determine the exact thresholdvalue (i.e., X12.9, X12.8,...).



Table 7. Regression Results† – Model 2

Dependent Variable: HHI Coefficients‡

Constant .531 (.0532)**

ConverterAdoptionRatiot–1 –.180 (.081)*

Var(Card_Price/MBt) 8.72e-07 (7.54e-07)

Condition Number 10.29

Log likelihood 431.2

†N = 428. ‡Standard errors in parentheses. *p < 5%;. **p < 1%.

ficant advantage in standards competition, and may lead to a“winner-take-all” outcome. However, Hypothesis 4 predictsthat when converters are available this competitive advantagewill be weakened and that market concentration will decreaseas the adoption of digital converters increases. We use thefollowing model to test Hypothesis 4 and to examine therelationship between the adoption of digital converters and theshift in market shares of the competing flash memory cardformats:

MarketConcentrationt = β0 + β1ConverterAdoptionRatiot–1

+ β2Var(CardPrice / MB)k,t + gk,t

[Model 2]

In Model 2, the dependent variable, market concentration, ismeasured by the Herfindahl–Hirschman Index (HHI), awidely used measure of competition in a market (Calkins1983). A larger HHI value indicates higher market concentra-tion and hence less intense competition. Note that in ourmodel HHI is computed as the sum of squares of the marketshares of the competing formats rather than brands as we areprimarily interested in competition among technology formatsrather than firms. To account for the fluctuation of the flashmemory card sales volume, we normalize our key indepen-dent variable, ConverterAdoption, by computing a newvariable, ConverterAdoptionRatio, as the cumulative numberof converters sold up to time t divided by the cumulativenumber of flash memory cards sold up to time t. To performthis analysis, we compute the HHI for each month–capacitypair, which results in 428 data panels. Next, we regress theHHI value in each panel against the lagged adoption ratio ofdigital converters of the corresponding panel, controlling forthe variance of flash memory card retail prices in each panel.26

In order to correct for price differences across differentmemory capacities, we calculate retail price as the averageprice per megabyte in each panel.

The new data set consists of panels spanning 12 capacitycategories and 44 consecutive months. As above, a Breusch–Pagan/Cook–Weisberg test and a Wooldridge test on theresiduals of the OLS regression confirm that both heteroske-dasticity and panel specific first-order autocorrelation (AR1)are present in our data. Therefore, as before, we adopt theGLS adjustments to correct for heteroskedasticity andautocorrelation.27

Table 7 presents the GLS regression results for Model 2. Thecoefficient estimate of the variable ConverterAdoptiont–1 isnegative and significant at the 5% level in both specifications,suggesting that market concentration decreases as the adop-tion of digital converters increases across different flashmemory card capacity categories. Hence Hypothesis 4 issupported.

Interpretation of Results

The main empirical question in this study is how conversiontechnologies affect standards competition in digital markets.Our findings provide insights that help to answer thisquestion.

First, we find that, consistent with both industry views and theacademic literature, the leading flash memory card formatsenjoy a product price premium that is mainly attributed to the

26This controls for variation in market concentration due to price variationthat is associated with other exogenous factors.

27We have also taken some steps to address the potential endogeneity inModel 2. First, we use the lagged term of converter adoption as our depen-dent variable as the present adoption of flash memory cards is less likely todrive the previous period adoption of multiformat converters. Next, we con-struct an instrumental variable using a similar approach as in Model 1. Wedifferenced the variable and used the residuals as the instrument and per-formed 2SLS. Following Baum et al. (2003), we tested for endogeneity inour augmented form using a generalized methods of moments (GMM)estimation. The results from the GMM estimation are consistent with thoseobtained from the OLS and GLS estimations. This provides confidence inour results against potential endogeneity concerns.



Figure 4. Price Premium of a Type of FlashCard at + or – 2 Standard Deviations of theMean Flash Card Installed Base

Figure 5. Price Premium of a Type of FlashCard at + or – 2 Standard Deviations of theMean Converter Adoption Level

format’s installed base. This provides support for Hypoth-esis 1.

Given this and the earlier arguments, one might expect adominant standard to emerge in this market. However, wealso find that the presence of digital converters offsets someof the impact of the installed base on product prices. Asshown in Table 6, the presence of converters in the flashmemory card market weakens the relationship between theinstalled base and price premia. At average levels of the flashmemory card installed base, the marginal impact of converterson price premia is negative, contradicting Hypothesis 2. Theimpact is positive only for flash memory card formats with aninstalled base below the average. For a flash memory cardformat with a smaller installed base (i.e., one standard devia-tion below the mean), a 1% increase in the adoption level ofdigital converters raises the flash card premium by an esti-mated $0.10. But, this price premium disappears when aformat’s installed base is close to the industry average.Intuitively, a converter serves as a tool for data exchangebetween devices using otherwise incompatible flash memoryformats. Such a converter is relatively more valuable for con-sumers who own a minority format as it allows them tocommunicate with consumers in a much larger network.Hence the utility gain, and consequently the willingness topay a higher price, is larger for consumers of the minorityformats than for those of the dominant format.

Of course, in addition to the differential impact of converterson consumers who belong to different networks, other factorsmay account for the lack of support for Hypothesis 2. For

example, flash memory card vendors can also profit from thesales of digital converters. If a vendor is engaged in the saleof both converters and flash cards, the vendor can theoreti-cally transfer some of the price premium from the flashmemory cards to the sales of digital converters and still profitoverall. A similar argument applies for vendors who do notproduce their own converters, but license a third party vendorto do so. In this case, the licensing fee could more than com-pensate for the loss due to flash memory card price reduction.

However, it is also important to note that, while the pricepremium of the leading formats is weakened with extensiveadoption of digital converters, the effects are not fully elimi-nated. This implies that a larger installed base is still a com-petitive advantage over other competing formats, and that thisadvantage is more significant as a format’s installed basegrows. All else being equal, a 1% increase in the installedbase of a flash card format gives rise to a $0.37 price premium(0.61% price increase) for a compatible flash memory card,28

whereas a 1% increase in other key product features such ascapacity, speed, and product reputation (review ratings) isassociated with a price increase of $0.28, $0.16, and $0.07,respectively. The comparison shows that network effects, asmeasured in this study, have a significant impact on the pricesof the flash memory cards. Moreover, it is also important tonote that the strength of network effects, as measured by theprice premium associated with the size of the installed base,

28Interestingly, this is a price premium of a similar order of magnitude(0.75% price increase) as that found in the microcomputer spreadsheet soft-ware market by Brynjolfsson and Kemerer (1996).



is sensitive to the adoption of digital converters. When thelevel of digital converters adoption is relatively low (i.e., onestandard deviation below the mean), the price premiumincreases by more than 100%, to $ 0.81. However, when theadoption of digital converters is high (i.e., one standarddeviation above the mean), the price premium almostdisappears.

Figures 4 and 5 illustrate the interaction between networkeffects (installed base) and the adoption of digital converters.In Figure 4, the x-axis denotes the adoption level of digitalconverters, and the y-axis represents the price premium of atype of flash card. The dashed, dotted, and solid lines repre-sent the price premium of a flash card format with a large (+2standard deviations), average, and small (–2 standard devia-tions) installed base, respectively. Figure 4 shows that theprice premium of a flash card format with a larger installedbase decreases as the adoption of digital converters increases,whereas the price premium of a flash card format with asmaller installed base increases as the adoption of digital con-verters increases. This supports Hypothesis 3 and suggeststhat there is a negative interaction effect between the installedbase of the flash card format and the sales of digitalconverters. In other words, the adoption of digital convertershas an opposite impact on the price premia of majority andminority flash memory card formats in our data.

A similar interaction is depicted in Figure 5, where the x-axisdenotes the installed base of a flash memory card format andthe y-axis represents the price premium of that type of flashmemory card. The dashed, dotted, and solid lines representthe price premium of a card format when the adoption ofdigital converters is high (+2 standard deviations), average,and low (–2 standard deviations), respectively. One can seethat network effects as measured here are present (indicatedby the upward slope of the price premium curve) only whenthe adoption of digital converters is below a certain level.When there is extensive adoption of digital converters (i.e.,above +2 standard deviations), our measure of network effectshas no impact on the market (the slope of the price premiumcurve is negative).

In addition to moderating the impact of installed base onproduct prices, our findings reveal that digital converters mayplay an important role in competition in the flash memorycard market. The results from Model 2 suggest that marketcompetition intensifies as converters are widely adopted, andthat first-mover advantage from installed base is relativelylow. This may heighten the attractiveness of these markets tonew entrants. Moreover, increasing converter penetrationreduces the competitive gap between various flash memorycard formats. This may cause buyers to focus on other pro-

duct attributes, in turn allowing competition to arise in otherdimensions, such as quality and performance. Producers ofleading formats may no longer be able to rely on a largeinstalled base to deter new entrants and to suppress compe-tition, as such a competitive advantage is likely to erode overtime as converters become widely available. In contrast to theself-reinforcing loop in classic network effects theory, in thepresence of digital converters, the larger the leading format’sinstalled base, the more such a benefit can be appropriatedamong consumers of the minority formats, creating an effectthat pushes the market away from high concentration. Undersuch a competitive environment, a standards competitioncharacterized with extensive adoption of converters is likelyto undergo a less predictable growth path.

Our results are also consistent with observed market behavior: converters have become increasingly important in the marketand have been disproportionately adopted by new entrants andminority formats. Sony, the major retailer of Memory Stickcards, has a worldwide initiative to promote its Memory Stickcard reader to be installed on a variety of laptops and desk-tops. Today, more and more PC manufacturers are includingflash memory card readers as a standard component on theirPCs, suggesting increasing consensus among different marketparticipants about the prospects for, and the importance of,digital conversion.

From a societal standpoint, our findings have important impli-cations for technology innovation and adoption. In many ITindustries, when the market cannot settle on an industry-widetechnology standard, both consumers and content providers(or application developers) may postpone their investmentsuntil the market is clear about which standard to adopt,resulting in uncertainty about the future of the technology (so-called “excess inertia”). However, if digital converters wereavailable to convert data between otherwise incompatibleformats, consumers may be more willing to embrace the newtechnology because the risk of being stranded would decline. Moreover, once any excess inertia among stragglers isovercome, technology adoption can be expected to accelerateand lead to a traditional evolutionary path.

Limitations and Future Research

Of course, as is generally true with empirical research, ourresults are subject to interpretation, and are limited to the dataavailable. The most obvious threat to the validity of the con-clusions reached here is the longitudinal nature of the data.Although this data set demonstrates the lack of an expectedwinner-take-all result during the time period of our study, it



is possible that the role of converters has been to only greatlydelay such an outcome, and not prevent it. Future researchcould be devoted to extending these results through additionalyears of data, to the degree that technological progress willpermit such comparisons.29

The current data also do not allow for a clear separation of thedegree to which the network effects as measured here aredirect network effects or indirect network effects. Althoughexamples of both have been suggested, it is unclear whetherone or the other dominates. Future research could focus ondeepening the analysis to collect data to disambiguate thesetwo effects.

It is also worth noting that, although we have performedvarious robustness tests to ensure that our model captures thedynamics in the flash memory card market, there are severaladditional factors that may be relevant. For example, despitethe superiority of the newer generation formats, the olderformats may still persist (with small and generally decliningmarket shares) due to incremental technical advances and thepresence of older model devices needing cards in these for-mats. Moreover, no single format dominates the other for-mats in terms of attributes such as form factor, transfer speed,upgradability, or cost, which may lead to a horizontally dif-ferentiated market. Parsing out the potential magnitude ofthese alternative effects could be interesting extensions of ourresearch that could be pursued in future studies.

The dynamic nature of the flash memory card market alsoraises several interesting questions for future research. Whenfirms can supply both flash memory cards and digital con-verters, their pricing strategies in both markets will beimportant to both researchers and practitioners. Moreover,although proprietary standards prevailed in the early stage ofthe flash memory card market, upon the advent of the digitalconverters, several proprietary standard owners began toreach cross-licensing agreements and promoted ease ofconversion between competing formats.30 This could be con-sidered a strategic move to take advantage of the introductionof digital converters and to cope with the perceived futurecompetition. Although we have demonstrated a possiblerationale behind such moves, the actual impact on firms’profits merits future empirical examination. Finally, both

social welfare and private surplus are likely to be affected bythe introduction of conversion technologies. Further studiesto quantify these impacts could provide important guidancefor policy makers concerned about the nature and conse-quences of new technology adoption in markets with networkeffects and digital conversion.

Conclusions

While the implications of network effects have been widelydiscussed in both the academic literature and in the popularpress, most illustrations of network effects are drawn fromexisting physical and analog environments. Our contention isthat the unique characteristics of digital environments mayalter some of the conventional wisdom about network effectsand their competitive implications. While managers havebeen taught to expect strong network effects to dominateplatform competition (e.g., VHS versus Blu-Ray versus HD-DVD), this wisdom may not serve them well in a digitalenvironment where content can be converted easily betweenstandards.

In this study, we illustrate some of these issues in the contextof the flash memory card market, where, in spite of apparentnetwork effects, there are multiple competing standards andlittle evidence of market consolidation during the time periodof the study. Specifically, we apply a modified hedonicregression to an extensive data set cataloging prices and salesof flash memory cards and flash memory converters. Ourfindings yield several important insights into the dynamics ofstandards competition in digital goods markets. First, exten-sive adoption of digital converters reduces the importance ofa format’s installed base, as seen by a reduced price premiumof the leading flash format. As a result, new formats are morelikely to attract customers than in the absence of such digitalconverters. Second, competition intensifies as the marketpower of the leading format is weakened, as reflected by adecreasing market concentration ratio with increasing con-verter adoption. These findings explain the seemingly coun-terintuitive trend of the lack of standards convergence cur-rently seen in the flash memory card market. Our findingsalso shed light on the likely evolution of standards compe-tition in other, similar digital product markets, such as thedigital media and image files markets. In these markets,various kinds of conversion software have emerged to facili-tate the conversion between the incompatible formats. At thesame time, we have seen increasing efforts from vendors ofdifferent media formats to promote such conversion. Theseefforts have significantly motivated consumers to adopt thenew media formats, leading to technological development inthese markets.

29As technologies evolve it may make it difficult to create “apples to apples”longitudinal data sets, as newer products of a flash memory card format maybe incompatible with older products of the same format.

30For example, SanDisk has been selling SanDisk-branded Memory Stickproducts and is a codeveloper with Sony of Memory Stick Pro, and Sony hassupplied a card reader on its laptops that can read SanDisk’s SD cards.



Acknowledgments

Financial support for this research was provided by the NETInstitute (Liu, Kemerer, and Smith), a National Science FoundationCAREER award IIS-0118767 (Smith), and a Katz Graduate Schoolof Business Dissertation Research Grant (Liu). The authors thankthe NPD research group for generously providing data on the flashmemory card market. We also thank Lorin Hitt, Jeffery Hu, ChrisForman, seminar participants at Carnegie Mellon University,Georgia Institute of Technology, INFORMS, the ICIS DoctoralConsortium, and the Workshop on Information Systems andEconomics (WISE), the senior and associate editors of this journal,and three anonymous referees for helpful comments on this research.

References

Aiken, L. S., and West, S. G. 1991. Multiple Regression: Testingand Interpreting Interactions, Newbury Park, CA: SagePublications.

Asvanund, A., Clay, K., Krishnan, R., and Smith, M. D. 2004. “AnEmpirical Analysis of Network Externalities in Peer-to-PeerMusic-Sharing Networks,” Information Systems Research (15:2)pp. 155-174.

Baum, C. F., Schaffer, M. E., and Stillman, S. 2003. “InstrumentalVariables and GMM: Estimation and Testing,” Working PaperNo.545, Department of Economics, Boston College.

Bresnahan, T. F. 2001. “Network Effects and Microsoft,” Depart-ment of Economics, Stanford University.

Brynjolfsson, E., and Kemerer, C. F. 1996. “Network Externalitiesin Microcomputer Software: An Econometric Analysis of theSpreadsheet Market,” Management Science (42:12), pp.1627-1647.

Calkins, S. 1983. “The New Merger Guidelines and theHerfindahl–Hirschman Index,” California Law Review (71:402),p. 28.

Chen, P. Y., and Forman, C. 2006. “Can Vendors InfluenceSwitching Costs and Compatibility in an Environment with OpenStandards?,” MIS Quarterly (30:SI), pp. 541-562.

Chen, P. Y., and Hitt, L. M. 2002. “Measuring Switching Costs andthe Determinants of Customer Retention in Internet-EnabledBusinesses: A Study of the Online Brokerage Industry,”Information Systems Research (13:3), pp. 255-274.

Choi, J. P. 1996. “Do Converters Facilitate the Transition to a NewIncompatible Technology? A Dynamic Analysis of Converters,”International Journal of Industrial Organization (14:6), pp.825-835.

Choi, J. P. 1997. “The Provision of (Two-Way) Converters in theTransition Process to a New Incompatible Technology,” Journalof Industrial Economics (45:2), pp. 139-153.

Chow, G. C. 1967. “Technical Change and the Demand for Com-puters,” American Economic Review (57:7), pp. 1117-1130.

Clements, M. T., and Ohashi, H. 2005. “Indirect Network Effectsand the Product Cycle: Video Games in the U.S., 1994-2002,”Journal of Industrial Economics (43:4), pp. 515-542.

Conner, K. R. 1995. “Obtaining Strategic Advantage from BeingImitated: When Can Encouraging ‘Clones’ Pay?,” ManagementScience (41:2), pp. 209-225.

Cusumano, M. A., Mylonadis, Y., and Rosenbloom, R. S. 1992.“Strategic Maneuvering and Mass-Market Dynamics—TheTriumph of VHS over Beta,” Business History Review (66:1), pp.51-94.

David, P. A., and Bunn, J. A. 1988. “The Economics of GatewayTechnologies and Network Evolution: Lessons from ElectricitySupply History,” Information Economics and Policy (3:2), pp.165-202.

de Vries, H. J., de Ruijter, J. P. M., and Argam, N. 2009.“Dominant Design or Multiple Designs: The Flash Memory CardCase,” Technical Report ERS-2009-032-LIS, Erasmus ResearchInstitute of Management (ERIM).

Dranove, D., and Gandal, N. 2003. “THE DVD vs. DIVX Stan-dard War: Empirical Evidence of Network Effects and Prean-nouncement Effects,” Journal of Economics & ManagementStrategy (12:3), pp. 363-386.

Economides, N. 1989. “Desirability of Compatibility in theAbsence of Network Externalities,” American Economic Review(79:5), pp. 1165-1181.

Economides, N. 1991. “Compatibility and the Creation of SharedNetworks,” in Electronic Services Networks: A Business andPublic Policy Challenge, M. Guerin-Calvert and S. Wildman(eds.), New York: Praeger Publishing Inc.

Economides, N. 1996. “The Economics of Networks,” Interna-tional Journal of Industrial Organization (14:6), pp. 673-699.

Economides, N., and Himmelberg, C. 1995. “Critical Mass andNetwork Size with Application to the US Fax Market,”Discussion Paper # EC-95-11, Stern School of Business, NewYork University.

Farrell, J., and Saloner, G. 1985. “Standardization, Compatibility,and Innovation,” RAND Journal of Economics (16:1), pp. 70-83.

Farrell, J., and Saloner, G. 1992. “Converters, Compatibility, andthe Control of Interfaces,” Journal of Industrial Economics(40:1), pp. 9-35.

Farrell, J., Shapiro, C., Nelson, R. R., and Noll, R. G. 1992.“Standard Setting in High-Definition Television,” BrookingsPapers on Economic Activity. Microeconomics.

Gallaugher, J. M., and Wang, Y.-M. 2002. “UnderstandingNetwork Effects in Software Markets: Evidence from WebServer Pricing,” MIS Quarterly (26:4), pp. 303-327.

Gandal, N. 1994. “Hedonic Price Indexes for Spreadsheets and anEmpirical Test for Network Externalities,” RAND Journal ofEconomics (25:1), pp. 160-170.

Gandal, N. 1995. “Competing Compatibility Standards and Net-work Externalities in the PC Software Market,” Review ofEconomics and Statistics (77:4), pp. 599-608.

Greene, W. H. 2003. Econometric Analysis, Upper Saddle River,NJ: Prentice Hall.

Greenstein, S. M. 1993. “Did Installed Base Give an IncumbentAny (Measurable) Advantages in Federal Computer Procure-ment?,” RAND Journal of Economics (24:1), pp. 19-39.

Hartman, R. S. 1989. “An Empirical Model of Product Design andPricing Strategy,” International Journal of Industrial Organi-zation (7), pp. 419-436.

Jaccard, J., Wan, C. K., and Turrisi, R. 1990. “The Detection andInterpretation of Interaction Effects Between ContinuousVariables in Multiple Regression,” Multivariate BehavioralResearch (25:4), pp. 467-478.



Katz, M. L., and Shapiro, C. 1985. “Network Externalities, Com-petition, and Compatibility,” American Economic Review (75:3),pp. 424-440.

Kauffman, R. J., McAndrews, J., and Wang, Y. 2000. “Opening the‘Black Box’ of Network Externalities in Network Adoption,”Information Systems Research (11:1), pp. 61-82.

Lee, J., Lee, J., and Lee, H. 2003. “Exploration and Exploitation inthe Presence of Network Externalities,” Management Science(49:4), pp. 553-570.

Liu, C. Z., Gal-Or, E., Kemerer, C. F., and Smith, M. D. 2011.“Compatibility and Proprietary Standards: The Impact of Con-version Technologies in IT-Markets with Network Effects,”Information Systems Research (22:1), pp. 188-207.

Matutes, C., and Regibeau, P. 1988. “‘Mix and Match’: ProductCompatibility Without Network Externalities,” RAND Journal ofEconomics (19:2), pp. 221-234.

McWilliams, G. 2003. “The Best Way to Solve Your MemoryProblem,” Wall Street Journal, March 27.

Mittal, N., and Nault, B. R. 2009. “Research Note—Investments inInformation Technology: Indirect Effects and InformationTechnology Intensity,” Information Systems Research (20:1), pp.140-154.

Nair, H., Chintagunta, P. K., and Dube, J.-P. H. 2004. “EmpiricalAnalysis of Indirect Network Effects in the Market for PersonalDigital Assistants,” Quantitative Marketing and Economics (2:1),pp. 23-58.

Park, S. 2004. “Quantitative Analysis of Network Externalities inCompeting Technologies: The VCR Case,” Review of Econo-mics & Statistics (86:4), pp. 937-945.

Shapiro, C., and Varian, H. R. 1999. Information Rules: AStrategic Guide to the Network Economy, Boston: HarvardBusiness School Press.

Song, J., and Walden, E. A. 2003. “Consumer Behavior in theAdoption of Peer-to-Peer Technologies: An Empirical Examina-tion of Information Cascades and Network Externalities,” inProceedings of the 9th Americas Conference on InformationSystems, Tampa, FL, August 4-6, pp. 1801-1810.

Song, J., and Walden, E. A. 2009. “Size Doesn’t Matter: NetworkExternalities vs. Information Cascades in the Adoption of LowCost Internet Technologies,” Working Paper, Rawls College ofBusiness Administration, Texas Tech University.

Suárez, F .F., and Utterback, J. M. 1995. “Dominant Designs andthe Survival of Firms,” Strategic Management Journal (16:6), pp.415-430.

Sundararajan, A. 2007. “Local Network Effects and ComplexNetwork Structure,” Center for Digital Economy ResearchWorking Paper, Stern School of Business, New York University.

Verbeek, M. 2004. A Guide to Modern Econometrics (2nd ed.),Hoboken, NJ: John Wiley & Sons, Inc.

Verbeek, M., and Nijman, T. 1996. “Incomplete Panels and Selec-tion Bias,” in The Econometrics of Panel Data: A Handbook ofthe Theory with Applications, L. Matyas and P. Sevestre (eds.),Dordrecht, Germany: Kluwer Academic Publishers, pp. 449-490.

Zhu, K., Kraemer, K. L., Gurbaxani, V., and Xu, X. 2006. “Migra-tion to Open-Standard Interorganizational Systems: NetworkEffects, Switching Costs and Path Dependency,” MIS Quarterly(30:SI), pp. 515-539.

About the Authors

Charles Zhechao Liu is an assistant professor of InformationSystems at the University of Texas at San Antonio. He received anM.A. in Economics from Tulane University and a Ph.D. in Manage-ment Information Systems from the Katz Graduate School of Busi-ness, University of Pittsburgh. His research interests include tech-nology adoption and diffusion, the economics of IS, and standardscompetition in IT markets. As a Ph.D. student, Charles won the2007 Net Institute Research Grant and was an ICIS DoctoralConsortium Fellow (2006). His research has been presented asconferences such as ICIS, INFORMS, and WISE, and has appearedin journals such as Information Systems Research and Communi-cations of AIS.

Chris F. Kemerer is the David M. Roderick Professor ofInformation Systems at the University of Pittsburgh and an AdjunctProfessor of Computer Science at Carnegie Mellon University.Previously, he was an Associate Professor at MIT’s Sloan School ofManagement. He has published more than 60 articles on informa-tion systems and software engineering in a variety of professionaland academic journals, including an MIS Quarterly Best Paperaward winner, as well as editing two books. He was recently namedto the ISI/Thomson Reuters list of Highly Cited Researcher inComputer Science, and has been selected as an InternationalAffiliate of King Abdul Aziz University in Saudi Arabia. Dr.Kemerer has held a variety of editorial positions at leading journalsand conferences, including Editor-in-Chief of Information SystemsResearch, Departmental Editor for Information Systems atManagement Science, Senior Editor for MIS Quarterly, and ProgramCo-chair for ICIS and WISE.

Sandra A. Slaughter (Ph.D., University of Minnesota), is the AltonM. Costley Chair and Professor of Information Technology Manage-ment at the Georgia Institute of Technology. Her thesis won firstplace in the doctoral dissertation competition held by the Inter-national Conference on Information Systems (ICIS) in 1995. Sincethen, she has gone on to publish more than 90 articles in leadingresearch journals, conference proceedings, and edited books. Herwork has received seven best paper awards at major conferences andthe best published paper award from Information Systems Research. Her research has been supported by grants from the NationalScience Foundation, the Department of Defense, and ResearchCenters at Georgia Tech, Carnegie Mellon University, and theUniversity of Minnesota. She was awarded the Xerox ResearchChair at Carnegie Mellon University in 1998. She currently servesas a departmental editor for Management Science (InformationSystems Department), and has served as a senior editor or associateeditor for other leading journals including MIS Quarterly, Informa-tion Systems Research, and others. In 2009, she served as ProgramCo-chair of ICIS.

Michael D. Smith is a Professor of Information Systems andMarketing at Carnegie Mellon University. He holds academicappointments at the School of Information Systems and Manage-ment and the Tepper School of Business. He received a Bachelorsof Science in Electrical Engineering (summa cum laude) and aMasters of Science in Telecommunications Science from the



University of Maryland, and a Ph.D. in Management Science fromthe Sloan School of Management at MIT. Professor Smith’sresearch uses economic and statistical techniques to analyze firmand consumer behavior in online markets. His research in this areahas been published in leading Management Science, Economics, andMarketing journals and covered by professional journals including

The Harvard Business Review and The Sloan Management Review.Professor Smith has also received several awards for his workincluding the National Science Foundation’s prestigious CAREERResearch Award, and was selected as one of the top 100 “emergingengineering leaders in the United States” by the National Academyof Engineering.

Appendix

Converter Use for Data Exchange Examples

A number of people like to view the video they took with their digital camcorders on large-size high definition TVs. To cater to theseconsumers, some manufacturers have developed TVs (or TV accessories such as DVD players) that are capable of playing digital video/photosstored on flash memory cards. For example, Panasonic’s VIERA plasma TVs have several models that come with a built-in SD card slot. TheseTVs enable consumers to play video stored on a SD card and record TV directly onto an SD card in the MPEG4 format. A consumer can thenreplay the recordings on the TV itself, or on any portable video player that supports the SD format. Other manufacturers such as Philips, Sharp,and Sony have also developed TVs that support a similar feature. If users have digital devices that support incompatible formats, then dataexchange between cards (through a converter) can occur in both ways.

(1) TV < SD card < PC < Converter < Digital Camcorder with a Memory stick, – or –(2) Portable video player (that supports memory stick) < memory stick < PC < Converter < SD card < TV (with recording capability)

Digital picture frames (a.k.a. digital photo viewers) are a popular device to display digital photos in a regular size photo frame. Some of thesedigital picture frames only support a certain number of flash memory formats (i.e., SD). If a consumer has pictures taken by a camera that isnot compatible with the formats supported by the digital picture frame (i.e., Memory Stick), then in order to display her pictures in the digitalpicture frame she will need to undergo the following conversion process:

Digital Picture Frame < SD card < PC < Converter < Memory Stick

Printing digital photos using a self-assisted kiosk in retail stores such as Wal-Mart, Target, and CVS, is a common consumer behavior. Thedigital printers in these kiosks do not always support the full array of flash memory formats (for example, many kiosks do not support MMC,xD, or SmartMedia cards). Suppose a consumer has her pictures stored on one of these less popular flash memory cards. She should then takethe following steps before she can print her photos:

Digital Printer < SD card (or a USB thumb drive) < PC < Converter < MMC card

Many GPS devices allow consumers to save the map file (or route information) to a flash memory card and this map file can be transferredto another GPS or a Smartphone that has the navigation function. If these devices support different flash memory formats, then a conversionis needed:

GPS1 (or Smartphone 1) < SD card < PC < Converter < MMC card < GPS2 (or Smartphone 1)

Another clearly unofficial but apparently popular use of the converters involves transferring games stored on different incompatible flashmemory cards and playing them on different game consoles. As we know, different video game consoles are not compatible with each otherand support different flash memory card formats as their storage media (i.e., SD card for Nintendo, Memory Stick for PSP). However, thereare game enthusiasts who have come up with ways to play the video games on the rival company’s previously incompatible game console (i.e.,with a “modded” Nintendo Wii it may be possible to run homebrew emulators from SD cards and to play Xbox games on a PSP). In this case,the following conversion process is required:

PSP < Memory Stick < PC < Converter < SD card < Xbox


Documents

TANDARDS COMPETITION IN THE PRESENCE OF …ckemerer/standards-competition-misq.pdf · importantly for our study—digital flash memory cards (e.g., Compact Flash, SmartMedia, Secure